Classification of Photographs based on Perceived Aesthetic QualityJeff Hwang, Sean Shi

Department of Electrical Engineering, Stanford University

Aesthetic Classification Feature Extraction

Methodology Experimental Results

Hue: count # of distinct hues Saturation: compute average saturation Contrast: variance of pixel intensity

Spatial Correlation of Features

Extract features from each tile in partitioned image. Allow machine learning algorithm to infer relationships between the tiles.

Feature Selection

Entropy: measure of simplicity Blur: variance of the Laplacian Detail: ratio of subject edges to pixels

Dataset Scraped 2300 images from photo.net, each photograph rated between 1 and 7. We only consider photographs rated below 4.3 or above 6.

K-fold Cross Validation Performance was measured using 10-fold cross validation. Balanced number of positive/negative examples used.

Classifier Tuning Selected regularization, gamma, and kernel parameters of SVM via grid search.

80.96% 26.78%

19.04% 73.22%

Predicted

ActualLabel

True

False True

False

SVM Gaussian Kernel, C=1, =0.0810-fold Cross Validation Success

1

0

1 0